GPU (ground power unit) parallel processing method for real-time detection of hyperspectral target

Abstract

The invention particularly relates to a GPU (ground power unit) parallel processing method for real-time detection of a hyperspectral target, which is realized on hardware of a graphics processor (GPUS) by combining with a hyperspectral real-time detection algorithm based on recursive analysis thinking. The invention comprises the following steps: (1) inputting hyperspectral data, initialized height, original data height, original data width, and original data wave section number; copying the initialized data from a host machine end to an equipment end; initializing hyperspectral data; copying current detection pixel element data from the host machine end to the equipment end; updating an inverse matrix of a related matrix by using wood bury identical equation; configuring multithreading on the GPU, and detecting the hyperspectral image by combining with an abnormal detection operator, and at last, outputting the detection result. The GPU parallel processing method fully utilizes the maxwell structural GPU to speed up the complex calculation; the GPU parallel realization of the hyperspectral target real-time detection algorithm has practicability and universality; the parallel processing method for real-time detection of the hyperspectral target can be flexibly applied to an airborne or satellite borne platform carried with an embedded system.

Description

technical field [0001] The invention specifically relates to a GPU parallel processing method for real-time detection of hyperspectral targets realized by hardware on graphics processors (GPUs) in combination with a hyperspectral real-time detection algorithm based on recursive analysis ideas. Background technique [0002] The emergence of hyperspectral remote sensing is a revolution in remote sensing technology. With the improvement of spectral resolution, it enables the detection of ground objects that cannot be effectively detected in multispectral remote sensing, so it has been widely used. In many practical situations, researchers always do not have enough prior knowledge to characterize the statistics of target classes, such as special species in agronomy, abnormal migration in ecology, rare minerals in geology, toxic waste in environmental monitoring Excretion and oil spills, vehicles or aircraft in the battlefield, cancer cells or tumors in medical diagnosis, etc. ...

AI Technical Summary

Problems solved by technology

However, most of the existing algorithms are relatively complex and have high computational complexity, and the computing platform is serially calculated, so it cannot realize real-time detection

[0004] On the other hand, with the rapid development of modern remote sensing technology, while hyperspectral remote sensing acquires more abundant ground object information, its large amount of data also brings enormous pressure to data storage, satellite downlink data transmission and subsequent processing.

Images